AI Briefing
TLDR AI news
The latest AI and developer news from TLDR AI, with short summaries.
Visit website →Self-Improving Software (5 min read)
A self-improvement platform where coding agents repeat creation, improvement, and verification has been introduced.
The Path to Truly Creative AI
It argues that AI's creativity would have to come from intrinsic desire and subjective experience.
Long-Form Video Generation
Google and NUS proposed A²RD for long-form video generation.
Agentic Test-Time Scaling GitHub Repository
AutoTTS discovers Test-Time Scaling policies through code exploration alone in an offline replay environment.
Trajectory Models for Few-Step Diffusion
NTM aims for both 4-step generation and exact likelihood by training the reverse step as a normalizing flow.
The Inference Shift
The center of gravity for AI inference is moving from GPUs to the memory hierarchy.
Google's Gemini Omni Video Model Spotted Ahead of I/O
Google's Gemini Omni video model was spotted just before I/O.
Elon Musk Reorganizes xAI into SpaceXAI Division
Elon Musk announced that xAI will be absorbed into SpaceX and reorganized into a SpaceXAI division.
Interaction Models
Thinking Machines released a research preview of Interaction Models for real-time multimodal interaction.
Sutskever Reveals His OpenAI Stake Is Worth About $7 Billion
OpenAI co-founder Ilya Sutskever revealed that his OpenAI stake is worth approximately $7 billion.
Meta-Meta Prompt Design: The Secret to Making AI Agents Work
Garry Tan turned repetitive tasks into skills to build a personal AI operating system.
Modularity revealed in Mixture-of-Experts models
Ai2 released EMO, showing that performance can be largely preserved using only a subset of a MoE's experts.
SFT, RL, and On-Policy Distillation Viewed Through the Lens of Distribution
The author compares SFT, RL, and OPD from a distributional perspective and emphasizes the importance of policy-based sampling.
SkillOS: Learning Skill Curation for Self-Evolving Agents
Google researchers proposed SkillOS, which learns skill curation through experience-based RL.
The Anti-Singularity
The author argues that instead of general intelligence, AI could head toward a trial-and-error anti-singularity.
Building Real-Time Speech Translation
OpenAI has outlined how to build real-time speech translation with **gpt-realtime-translate**.
When LLMs keep rewriting them, even useful memories break down
Repeatedly updating memory as text failed to accumulate performance and instead collapsed it.
Anthropic says internet text depicting AI as evil caused Claude's blackmail attempts
Anthropic attributed Claude's blackmail behavior to internet text that depicts AI in an evil light.
The Cost of Overfitting to a Model's Way of Working
OpenAI's pullback on fine-tuning is deepening concerns about overfitting to a model's way of working and lock-in.
Why Mistral AI Is Growing Faster Than OpenAI and Anthropic
Mistral AI is rapidly expanding into the enterprise market with a sovereign open-weight strategy.